Best OpenClaw Hosting in 2026: 8 VPS Picks + How to Set It Up

For most people, the best OpenClaw hosting is a small Linux VPS you control with at least 4GB of RAM, and the fastest way there is Hostinger’s one-click OpenClaw template at $6.49/mo. Want the most RAM per dollar? Contabo gives you 8GB for about $6.40. Never want to touch a terminal? xCloud deploys it in minutes for $9.99/mo flat. And if your budget is $0, Oracle Cloud’s Always Free ARM tier still runs it. But one rule matters more than the host you pick: never expose OpenClaw’s dashboard port to the public internet. Researchers found more than 42,000 instances left wide open. Don’t be one of them.

OpenClaw is the open-source AI agent everyone suddenly wants to self-host. Austrian developer Peter Steinberger shipped it in late 2025 as Clawdbot, renamed it Moltbot in January 2026 after a trademark nudge from Anthropic, then settled on OpenClaw days later. The name kept changing; what it does didn’t. It sits on your own server and connects your messaging apps, WhatsApp, Telegram, Discord, Slack, Signal, iMessage, to an always-on AI assistant that can browse the web, run commands, manage files, and work while you sleep. You message it like a contact. It does real work. Your data stays on your box.

The catch is that “self-hosted” means you need a host. And the host you choose decides whether this is a five-minute setup or a weekend of SSH debugging, whether it costs $0 or $25 a month, and whether it stays online when your laptop goes to sleep. So let’s pick the right one.

Proof and last verified
Tested by	Gaurav Tiwari
Hands-on	I self-host OpenClaw via Docker in cloud-model mode
How I compared	7 hosts on RAM, entry price, and OpenClaw deploy path
Pricing checked	June 26, 2026
Best for	Developers and power users who want a private, always-on AI agent
Avoid if	You won’t keep it updated and secured, or you want a big local model on a cheap VPS
What changed	First publish
May change soon	Hostinger and HostArmada intro pricing, and Oracle’s free-tier ARM allowance

OpenClaw hosting at a glance: quick picks

Short on time? Here’s the one-line version for each type of person.

If you want…	My pick	Entry price
The easiest setup, running today	Hostinger VPS (one-click template)	$6.49/mo
Zero terminal, fully managed	xCloud	$9.99/mo flat
The most RAM per dollar	Contabo (8GB at entry)	~$6.40/mo
A hardened deploy for business use	DigitalOcean 1-click	$24/mo
Cheap entry with backups built in	HostArmada (Fusion plan)	$10.74/mo
The lowest yearly bill	RackNerd (manual Docker)	$32.49/yr
To run it free while testing	Oracle Cloud Always Free	$0

Full comparison: 8 OpenClaw hosts side by side

Every price below is the entry tier. The RAM column is what you actually get at that price, because RAM is the one spec that decides whether OpenClaw runs smoothly. More on why in a second.

Host	Entry price	Entry RAM	One-click OpenClaw?	Best for
Hostinger VPS	$6.49/mo*	4GB	Yes, dedicated template	Best all-round
xCloud	$9.99/mo flat	4GB	Yes, managed, no terminal	Hands-off setup
Contabo	~$6.40/mo	8GB	Yes, free add-on	Most RAM per dollar
DigitalOcean	$12/mo (rec. $24)	2GB (rec. 4GB)	Yes, hardened Marketplace	Production use
HostArmada	$3.69/mo (use Fusion)	1GB (Fusion 8GB)	Preinstalled image	Cheap entry + backups
RackNerd	$32.49/yr*	3.5GB (4GB $59.99/yr)	No, manual Docker	Cheapest annual bill
Cloudways	$11/mo	1GB	No, manual Docker	Existing Cloudways users
Oracle Cloud Free	$0	12GB (ARM)	No, manual ARM	Free testing

What OpenClaw actually needs from a host

Here’s the thing most “OpenClaw hosting” pages get wrong: they sell you a giant server you don’t need.

OpenClaw itself is light. The gateway, the part that runs on your VPS and routes messages, is just a Node.js process. In the normal “cloud-model” setup, where you hand it a Claude, GPT, or Gemini API key, the heavy thinking happens on Anthropic’s or OpenAI’s servers, not yours. Your box is only the messenger. Here’s what that actually demands from an OpenClaw hosting plan:

• RAM: 4GB is a comfortable floor; 2GB runs one or two channels.
• Storage: months of chat history is only 100 to 500MB.
• CPU: the Node.js gateway barely touches it.
• Who this fits: roughly 90% of people self-hosting OpenClaw.

The other mode is where it gets hungry. If you run the model locally with Ollama, so you pay $0 in API fees and your prompts never leave your server, the model itself has to fit in RAM. An 8-billion-parameter model at Q4 quantization needs roughly 8GB on its own, and that’s before the gateway. For local models you want 16GB and ideally a GPU. Different game entirely.

The buying rule, then, is simple:

• Using a cloud API key (Claude/GPT/Gemini)? 4GB RAM is the comfortable floor. 2GB works for one or two channels.
• Running a local model with Ollama? 16GB minimum, and budget for a GPU instance.

If you’re new to renting servers, my tested web hosting picks and the Hetzner vs Vultr vs RackNerd cost breakdown cover the fundamentals of choosing a VPS without overpaying. Everything here assumes the cloud-model setup unless I say otherwise.

How I picked these 8

I run OpenClaw on my own VPS, the standard Docker install in cloud-model mode, so this isn’t a spec-sheet roundup written from a press release. I compared the eight OpenClaw hosting providers below on four things that actually change your experience:

1. The deploy path. Is there a real one-click OpenClaw template, or are you doing it by hand in Docker? Both are fine. The page should tell you which you’re getting.
2. RAM per dollar at the entry tier. Because that’s the spec that matters and the one hosts love to skimp on.
3. The safety net. Money-back window, free trial, or free credit. Self-hosting an agent is the kind of thing you want to be able to walk away from.
4. Honest fit. A couple of these hosts are on every “OpenClaw hosting” list and shouldn’t be. I’ll tell you which.

Prices were checked on June 26, 2026. Hosting prices rot fast, so trust the “last verified” date, not the calendar.

The 8 best OpenClaw hosts

1. Hostinger VPS, the best all-round pick

Hostinger gets the top spot because it’s the only host here that pairs a true one-click OpenClaw template with a real 30-day money-back guarantee and the lowest entry price for 4GB of RAM.

You pick the OpenClaw template, the KVM 1 plan gives you 1 vCPU, 4GB RAM, 50GB NVMe and 4TB of bandwidth on AMD EPYC hardware, and it provisions OpenClaw for you. There’s a Docker manager and one-click updates in the panel, so you’re not SSHing in to pull new images. For the cloud-model setup, KVM 1 at $6.49/mo is genuinely all you need. If you’ll run several busy channels, the KVM 2 plan (“Most Popular”) doubles you to 2 vCPU and 8GB for $8.99/mo.

One honest downside: that $6.49 is the intro price on a long, roughly two-year prepay, and it renews near $11.99. Budget for the renewal, or treat the first term as a cheap, refundable trial.

Best for: Almost everyone. First-timers who want OpenClaw running today without learning Docker. If you’ve outgrown shared hosting and want a private AI agent without a learning curve, Hostinger is the easy answer.

Get the Hostinger OpenClaw VPS

2. xCloud, for people who never want to see a terminal

xCloud is the odd one out here, and that’s the point. It’s not a VPS host. It’s a managed server panel that deploys and runs OpenClaw for you, live in minutes, no Docker, no SSH, no config files. You sign up, connect a Telegram bot token, and you’re messaging your agent.

The price is refreshingly honest: $9.99/mo, all-in, flat. No intro-then-renewal trick. What you’re paying for is the management layer: free SSL, automatic provisioning, updates handled. And because xCloud is a panel, you can point it at your own cloud account, DigitalOcean, Vultr, AWS, Google Cloud, and have it manage OpenClaw on infrastructure you already own.

One honest downside: you’re paying a management fee on top of whatever the server costs. If you’re comfortable in a terminal, you can get the same result cheaper. xCloud’s pitch is that your time is worth more than the difference.

Best for: Non-developers, and developers who’d rather pay $10 than babysit another box.

Try xCloud’s one-click OpenClaw

3. Contabo, the most RAM per dollar

If your priority is raw resources for the money, nothing on this list touches Contabo. The entry Cloud VPS 10 is 4 vCPU, 8GB RAM, and 75GB of NVMe for about €5.50 (roughly $6.40) a month. That’s double the RAM of most $6 VPS plans, which gives you real headroom if you want to run a few channels, keep long conversation history, or experiment with a small local model later.

Contabo confirms a free one-click OpenClaw add-on that automates the full setup, so you’re not stuck doing it manually despite the budget price.

One honest downside: there’s no money-back guarantee, you may hit a one-time setup fee on the cheapest term, and Contabo’s provisioning and support have historically been slower than the premium hosts. You trade a little polish for a lot of RAM.

Best for: Budget-conscious tinkerers who want maximum headroom and don’t mind a rougher edge. This is also the cheapest sane path toward eventually running a local model. For more cheap-server options in this vein, see my cheap dedicated and VPS hosting under $60 roundup.

See Contabo’s OpenClaw VPS

4. DigitalOcean, the most production-hardened deploy

DigitalOcean is where I’d put OpenClaw if I were running it for a business and uptime mattered. Its Marketplace one-click app isn’t just an installer, it’s security-hardened out of the box: OpenClaw on Ubuntu 24.04 with Node 22, Docker sandboxing, rate-limited firewall rules, non-root execution, and the gateway token already set. That’s the configuration most people forget to do themselves, done for you.

The droplet that DigitalOcean’s own image recommends is the 4GB tier (2 vCPU, 80GB) at $24/mo, though deployment technically starts at the $12/mo 2GB droplet. New accounts usually get $200 in credit for the first 60 days, so you can run it free for two months and decide.

One honest downside: it’s the priciest per gigabyte of RAM in this budget group, and there’s no money-back guarantee once the credit runs out.

Best for: Developers and teams who want a real, hardened, documented deploy and will use it seriously. I already use DigitalOcean across projects, and the DigitalOcean hosting setup work I do for clients is built on exactly this kind of droplet.

Deploy OpenClaw on DigitalOcean

5. HostArmada, cheapest door in with backups built in

HostArmada has the lowest entry price on the list, the Spark plan at $3.69/mo, and its VPS images ship with OpenClaw preinstalled and ready to run. Every plan includes root access, a dedicated IPv4, DDoS protection, automated backups and snapshots, across data centers in London, Frankfurt, Dallas, Montreal, Singapore, and Sydney.

But read the specs before you click Spark. At 1 vCPU and 1GB RAM, Spark is below OpenClaw’s comfortable floor. The plan you actually want is Fusion, 4 cores, 8GB RAM, 160GB NVMe, at $10.74/mo, which clears the bar with room to spare and still includes the snapshots.

One honest downside: the VPS money-back window is only 7 days (the famous 45-day guarantee is shared hosting only), and monthly-billed VPS plans aren’t refundable. Less room to change your mind.

Best for: People who want OpenClaw preinstalled with proper backups and snapshots, and who’ll spend the extra few dollars on Fusion instead of being tempted by Spark.

Compare HostArmada’s OpenClaw plans

6. RackNerd, the cheapest annual bill for OpenClaw

If your goal is the lowest yearly cost and you’re fine doing a ten-minute manual install, RackNerd is hard to argue with. It’s a no-frills KVM VPS provider with aggressive, price-locked-for-life annual deals across 21 data centers, and it’s the host I lean on for cheap staging boxes.

For OpenClaw in cloud-model mode, the plan that clears the 4GB floor is a 4GB RAM, 3 vCPU, 60GB SSD box at $59.99/year, about $5/mo billed annually. Want to spend even less? The popular 3.5GB plan (2 vCPU, 65GB SSD, 7TB bandwidth) is $32.49/year, just under the comfortable floor but fine for a channel or two. There’s no one-click OpenClaw template here, so you’ll run the Docker setup yourself, which takes about ten minutes on any host.

One honest downside: RackNerd is annual-prepay by design, the headline prices disappear on monthly billing, support is budget-tier, and there’s no real money-back guarantee. It’s a commit-for-the-year choice. You trade hand-holding for the lowest sticker on the list.

Best for: Cost-focused self-hosters who want the cheapest yearly OpenClaw bill and don’t mind the manual Docker route. If you’re weighing budget VPS providers, my Hetzner vs Vultr vs RackNerd breakdown shows where each one actually wins.

See RackNerd’s KVM VPS deals

7. Cloudways, only if you already live there

I’m including Cloudways because it’s on every OpenClaw list, and I want to be straight with you: it’s the weakest fit here.

Cloudways is a brilliant managed panel for WordPress, Laravel, and PHP apps, sitting on top of DigitalOcean, Vultr, Linode, AWS, or GCP. Entry is a 1GB DigitalOcean server at $11/mo ($8.25 on annual). But there’s no OpenClaw template. You’d be running it manually inside Docker on a platform that isn’t designed around arbitrary self-hosted agents. It works, but you get none of the one-click convenience that makes the other picks worth it.

One honest downside: the whole reason to use Cloudways, the managed app experience, doesn’t extend to OpenClaw. You pay the management margin and still do the work yourself.

Best for: People already running sites on Cloudways who want OpenClaw on the same bill and don’t mind a manual Docker setup. Everyone else, pick something above. If you’re shopping Cloudways generally, my Cloudways alternatives comparison is the better starting point.

View Cloudways pricing

8. Oracle Cloud Always Free, the genuine $0 option

Yes, you can run OpenClaw for free, permanently, and it’s not a trick. Oracle Cloud’s Always Free ARM (Ampere A1) tier is real and it’s over-spec for OpenClaw in cloud-model mode.

One important 2026 correction, though, because most guides still get this wrong: as of June 2026, Oracle cut the free ARM allowance for new free accounts from 4 OCPU / 24GB down to 2 OCPU / 12GB. Still plenty for OpenClaw, but no longer the 24GB monster the old tutorials promise.

One honest downside: there’s no one-click template (it’s a manual ARM Docker install), the free ARM instances are notoriously hard to actually grab (“out of capacity” errors are constant), and free accounts carry a small risk of reclamation. For an always-on agent, “free but might vanish” is a real tradeoff.

Best for: Hobbyists and testers who want $0, enjoy the setup challenge, and won’t cry if the instance hiccups. Not for anything you depend on.

Start with Oracle Cloud Free Tier

How to set up OpenClaw on your VPS (the Docker way)

One-click templates are great, but the manual Docker install takes ten minutes and works on any host, Contabo, Oracle, a raw droplet, anything. Here’s the path I actually use. The official docs are the source of truth; this is the short version with the gotchas marked.

Before you start, you’ll need:

• A Linux VPS, Ubuntu 24.04 is the safe choice
• Docker installed and running
• One AI provider API key (Anthropic, OpenAI, or Google)
• Node 24 recommended, or 22.19+ if you install the CLI directly instead of using Docker

Step 1, get on the box and update it. SSH in, run your updates, and make sure Docker is installed and running. Confirm your SSH key login works *before* you disable password login, locking yourself out of a fresh VPS is the most common self-inflicted wound here.

Step 2, run the Docker setup script. From the OpenClaw repo root, the canonical install is one command:

./scripts/docker/setup.sh

That script builds the image, walks you through onboarding (it’ll ask for your provider API key), generates a gateway token and writes it to your .env file, and starts the gateway with Docker Compose. If you’d rather pull the prebuilt image, set OPENCLAW_IMAGE="ghcr.io/openclaw/openclaw:latest" first, then run the same script.

Step 3, open the dashboard, locally. The gateway serves its web dashboard on port 18789. Do not open that port to the world. Reach it over an SSH tunnel from your laptop instead:

ssh -L 18789:127.0.0.1:18789 you@your-server-ip

Then visit http://127.0.0.1:18789/ in your browser and paste the gateway token from your .env. We’ll come back to why this step is non-negotiable.

Step 4, connect a messaging channel. This is the fun part. To link WhatsApp, run the channel login, scan the QR code with the phone you want the assistant to live on, and start the gateway:

openclaw channels login
openclaw gateway --port 18789

In your config (~/.openclaw/openclaw.json, it’s JSON5, so comments are allowed), lock the channel down to your own number so nobody else can command your agent:

{
  gateway: { mode: "local" },
  channels: { whatsapp: { allowFrom: ["+15555550123"] } },
}

Step 5, pick your model. Set it in the config as provider/model-id, for example a current Claude or GPT flagship. Check the provider’s current model name at setup time rather than copying an ID from an old tutorial, they change.

That’s it. Message your number and the agent answers. To update later, pull the latest image and restart the container, or set up Watchtower to do it automatically.

Don’t skip this: the security part everyone gets wrong

Here’s the uncomfortable truth about self-hosting OpenClaw. It can run shell commands and read and write files with whatever permissions you give it. That’s the whole point, and it’s also the whole risk. If the dashboard is exposed, or a dependency is compromised, an attacker inherits those same powers on your server.

This isn’t hypothetical. In early 2026, CrowdStrike researchers found more than 42,000 OpenClaw instances exposed on the public internet, and roughly 63% were running with the gateway open and no authentication at all. There was a real bug behind some of it too: CVE-2026-25253, a cross-site WebSocket hijacking flaw where simply visiting a malicious page could hand an attacker control of an exposed agent. It was patched in OpenClaw 2026.1.29. So it’s two rules, not one: keep the dashboard off the public internet, and keep OpenClaw updated.

So, the rules:

• Bind the gateway to localhost (127.0.0.1), never 0.0.0.0. Access it through an SSH tunnel, Tailscale, or a reverse proxy with real HTTPS and auth. Never raw on a public port.
• Keep the gateway token secret. It’s in your .env. Treat it like a password.
• Lock every channel to an allow-list. Anyone who gets into your linked WhatsApp or Telegram can command your agent. The allowFrom list is your gate.
• Run as a non-root user. The Docker image already does this (it runs as node, drops dangerous capabilities, sets no-new-privileges). Don’t undo that.
• Firewall down to 22, 80, 443. If you run local Ollama, block its port 11434 from the internet, it has no built-in auth at all.
• Leave the destructive-command confirmations on. Don’t auto-approve rm, sudo, or chmod. Let the agent ask.
• Keep OpenClaw patched. The CVE-2026-25253 fix shipped in 2026.1.29. Pull new images regularly, or let Watchtower do it for you.

If you do nothing else, do the first one. Loopback plus a tunnel is the difference between a private assistant and a public backdoor.

The cost nobody puts on the comparison page

The VPS is the cheap part. With a cloud model, your real bill is the API tokens, and an autonomous agent that loops while you sleep can run them up fast. There’s no spend cap by default.

A $6 VPS plus a chatty Claude or GPT key can easily cost more in tokens than in hosting. Before you turn an agent loose on autonomous tasks, understand what you’re paying per million tokens. My ChatGPT API pricing breakdown walks through real model costs and how to keep them down, and it applies directly to whatever model you point OpenClaw at.

This is also the real argument for the local-model route on a 16GB box: the server costs more, but the marginal cost of every message drops to electricity. For heavy, always-on use, that math flips. For occasional use, cloud models win easily.

Who should not self-host OpenClaw

Honesty time. Self-hosting OpenClaw is not for everyone, and the hosts above can’t fix that.

Skip it if you’re not comfortable with a terminal and you don’t want to pay for a managed option like xCloud. Skip it if you won’t keep the thing updated, an autonomous agent with shell access and stale dependencies is a liability. And skip it if you were hoping to run a big local model on a cheap VPS, that’s a GPU-and-16GB conversation, not a $6 droplet.

For the cloud-model setup, though, with the security basics in place? It’s one of the more genuinely useful things you can put on a server right now. If you’re still exploring what these agents can do before committing, my roundup of the best AI chatbots like ChatGPT is a lower-stakes place to start.

The verdict

If you want one answer: Hostinger’s one-click OpenClaw VPS at $6.49/mo is the best OpenClaw hosting choice for almost everyone, real template, 4GB RAM, 30-day refund. Want the most hardware for your money and don’t mind a rougher ride? Contabo’s 8GB for $6.40 is unbeatable. Refuse to touch a terminal? xCloud at $9.99 flat does it all for you. Running it for a business? DigitalOcean’s hardened one-click is the grown-up choice. And if you want to prove it works for $0 first, Oracle’s free tier still delivers.

But notice what the whole list has in common. The host is the easy decision. The setup is ten minutes. The thing that actually separates a useful private assistant from a cautionary tale is whether you bound that dashboard to localhost. Pick any host above. Then do that one thing right.

Frequently asked questions

Disclaimer: This site is reader-supported. If you buy through some links, I may earn a small commission at no extra cost to you. I only recommend tools I trust and would use myself. Your support helps keep gauravtiwari.org free and focused on real-world advice. Thanks. - Gaurav Tiwari